RAI output deflection error

Thank you for sharing your experience with Amazon Nova-Pro and the RAI (Responsible AI) output deflection error you encountered. You're correct that this type of error occurs when the generated output is determined to be misaligned with AWS's Responsible AI policy.

The error message "The generated text has been blocked by our content filters" is indeed a standard response for RAI output deflection errors in text generation. It's important to note that these filters are designed to ensure responsible and ethical use of AI technology.

While it may seem that the detection is overly sensitive, especially when compared to other models like those from Meta or Anthropic, it's worth considering that different companies may have varying standards and policies for content filtering. AWS has implemented these measures as part of their commitment to responsible AI use.

If you believe that your input and the expected output should not violate any content policies, you might want to consider the following:

1. Review your prompts: Sometimes, slight adjustments to how you phrase your requests can help avoid triggering content filters.

2. Examine the content: If you're converting PDF to text, there might be elements in the PDF that are triggering the filters, even if they're not immediately apparent.

3. Break down larger requests: If possible, try processing smaller sections of the PDF separately to identify if specific parts are causing the issue.

4. Contact AWS support: If you believe this is a false positive, you may want to reach out to AWS support for further clarification or to provide feedback on the sensitivity of the filters.

It's good that you have alternatives like Meta's model to fall back on. Your report of this experience is valuable for the community, as it helps others who might encounter similar situations understand what's happening and consider alternative approaches.
Sources
Error handling - Amazon Nova
Responsible use - Amazon Nova